Real-time update of calibration model for better monitoring of batch processes using spectroscopy

In order to reduce the large calibration matrix usually required for calibrating multiwavelength optical sensors, a simple algorithm based on the addn. in process of new stds. is proposed. A small calibration model, based on 14 stds., is periodically updated by spectra collected online during fermn. operation. Concns. related to these spectra are reconciled into best-estd. values, by considering carbon and oxygen balances. Using this method, fructose, acetate, and gluconacetan were monitored during batch fermns. of Gluconacetobacter xylinus 12281 using mid-IR spectroscopy. It is shown that this algorithm compensates for noncalibrated events such as prodn. or consumption of byproducts. The std. error of prediction (SEP) values were 0.99, 0.10, and 0.90 g/L for fructose, acetate, and gluconacetan, resp. By contrast, without an updating of the calibration model, the SEP values were 2.46, 0.92, and 1.04 g/L for fructose, acetate, and gluconacetan, resp. Using only 14 stds., it was therefore possible to approach the performance of an 88-std.-based calibration model having SEP values of 1.11, 0.37, and 0.79 g/L for fructose, acetate, and gluconacetan, resp. Therefore, the proposed algorithm is a valuable approach to reduce the calibration time of multiwavelength optical sensors. [on SciFinder (R)]